Segmental Semi-Markov Models for Change-Point Detection with Applications to Semiconductor Manufacturing

We formulate the problem of change-point detection in a segmental semi-Markov model framework where a change-point corresponds to state switching. The semi-Markov part of the model allows us to incorporate prior knowledge about the time of change in a Bayesian manner. The segmental part of the model allows exible modeling of the data within individual segments, e.g., as linear, quadratic, or other regression functions. This segmental semi-Markov model is an extension of the standard hidden Markov model (HMM), from which learning and inference algorithms are extended. Results on both simulated and real data from semiconductor manufacturing illustrate the exibility and accuracy of the proposed framework. 200 210 220 230 240 250 260 4500 5000 5500 6000 6500 7000 Time Y Figure 1: An illustrative example of a change-point detection problem from semiconductor manufacturing.